Recognition of Cheating Behavior in Examination Room Based on Deep Learning

In order to ensure the fairness of the examination and solve the problems that the traditional electronic invigilator system cannot automatically analyze the surveillance video and the video censors suffer from high labor intensity, this work presents a method for analyzing cheating behaviors of candidates based on deep learning technology. By analyzing the action of examinee in a single frame of the surveillance video, the traditional object detection algorithm YOLO is combined with the human posture estimation project OpenPose to extract the position information of the candidates, and label the candidates who are suspected of cheating. Due to the particularity of cheating behavior, a dataset for deep learning training is constructed, which marks two kinds of human behavior in the examination scene, including two types of cheating behavior: peeping and passing notes. At the same time, in order to improve the detection speed, the inter-frame difference method is used to extract the key frames, and finally both the accuracy on the test set and the detection speed can reach a better result.